Probabilistic modeling of influenza pandemics insurance implications Robert MuirWood Chief Research

1
Probabilistic modeling of influenzapandemics
insurance implications Robert Muir-Wood Chief
Research Officer Risk Management Solutions
Presentation to CEA, Stockholm ? June 15, 2007
2
A Science-based Pandemic Model

• Client demand led to RMS developing a
  probabilistic pandemic model to assess potential
  payouts for life insurers and reinsurers, and to
  inform other risk management decisions
• Model draws on the substantial body of medical
  research on the epidemiology of infectious
  disease, including state-of-the-art models of
  disease spread and containment
• Models uses published science of virology,
  clinical data and medical trials
• Historical research provides qualitative and
  quantitative data that give long term validation
  perspectives
• Science-based models provide a more comprehensive
  analysis than actuarial models

3
RMS Pandemic Impact Analysis for Insurance
Companies

• A. Company-Specific Losses
• Numbers of claims and payout values to insurance
  policy-holders for death or illness
• Life insurance Individual, Group, etc.
• Health insurance
• Other insurance lines offering coverage for death
  or injury
• Non-Life lines of coverage
• B. Investment Strategy
• Link between severity of outbreak and likely
  impacts on investments (and other economic
  consequences)
• C. Business Continuity Planning
• Insights from the modeling and potential
  scenarios to assist with preparedness plans to
  manage business continuity during a pandemic,
  including modeled analysis of operational
  productivity losses through staff sickness and
  office closures

4
RMS Pandemic Influenza Advisors Network

• Neil M. Ferguson, D.Phil., O.B.E., Professor of
  Mathematical Biology, Imperial College, London
• Leading authority on influenza pandemic advisor
  to World Health Organization and British
  Government
• Member of the 16 person World Health Organization
  Pandemic Task Force
• Marc Lipsitch, Associate Professor of
  Epidemiology, Harvard School of Public Health
• Highly-acclaimed research into the threat and
  spread of pandemic influenza
• Member of the 16 person World Health Organization
  Pandemic Task Force
• Eric K. Noji, Johns Hopkins, Bloomberg School of
  Public Health
• Chief of the Epidemiology, Surveillance and
  Emergency Response at the Centers for Disease
  Control Prevention in Atlanta
• White House advisor and responsible for U.S.
  National Pharmaceutical Stockpile

5
RMS Stochastic Model Framework
An Event Tree to produce 1,890 probability-weighte
d scenarios
Example
Probability of Pandemic
1
Probability of antigenic shift occurring
42 permutations of infectiousness and
virulence e.g. R0 of 2.25, Initial Mortality per
case of 2.5
Infectiousness and Lethality
2
Demographic Impact
3
3 Age mortality distributions Seasonal flul,
Cytokine Storm, Flat
5 Regions where the outbreak may begin e.g.
Region 1 Southeast Asia 'Under-developed'
Location of Outbreak
4
Vaccine Production
3 assumption sets about vaccine production e.g.
Optimistic 3 months 100m doses a month, 90
effective
5
Country-specific measures e.g. US National
Pandemic Preparedness Plan
National Counter-Measures
6
Three year lifecycle Years T0, T1, T2
Pandemic Lifecycle
7
6
Attempts to Contain Spread of Avian Flu

• Traffic in poultry by humans is now thought more
  responsible for long-distance spread than wild
  bird migration
• Many countries have now brought in new measures
  to control bird imports
• American continents still free of H5N1
• Containment measures for infected bird
  populations have been aggressive
• 200 million birds destroyed worldwide since 2004
• Feared explosion in bird infections appears to
  have been prevented

Apr 2007
Human cases since 2004 288 Deaths 170
7
The Characteristics of an Infectious Disease
30
Black Death England, winter of 1349
20
H5N1 caught from birds
10
20
5
Death Rate in Infected Cases
Reference Mortality Total death rate in a disease
cycle to the entire population, without
intervention
3
2
1
10
0.5
0.2
0.1
1918 flu epidemic
1957 flu epidemic
1968 flu epidemic
0
Average flu season, U.S.
0.0
10.0
20.0
30.0
40.0
50.0
60.0
of Population Infected
R0 Initial Reproductive Number
1.5
2.0
2.5
3.0
4.0
0.5
1.0
3.5
8
Infectiousness Lethality of Pandemic Viruses
Representative Pandemic Scenarios
Pandemic Severity
B
C
D
E
F
G
50
30
20
Niger Virus
10
Initial Deaths per Case
7.5
5
3.5
Hanoi Flu
British Influenza
1918
1918
2.5
2.0
1
0.75
Turkish Flu
1957
0.5
0.2
Normal
1968
0.1
0
1.0
1.5
2.0
2.5
3.0
3.5
4.0
1.25
1.75
2.25
2.75
3.25
3.75
R0 Initial Reproductive Number
9
Pandemic Planning Scenarios
Fictitious illustrations of potential influenza
pandemics
Infectiousness (Speed of spread)
Pathogenicity (lethality)
Illness Demographic (Ages worst hit)
Pandemic Severity
British Influenza Mild but likely pandemic H1N1
drift, Similar to 1968 pandemic
Slow spread R01.75
Low Initial DpC 0.5
Old and young worst affected
471 C2
Moderate Slow
Turkish Flu Moderate pandemic H5N1 shift, More
severe than 1957 pandemic
V. Rapid spread R03.25
Moderate Initial DpC 0.5
All ages equally affected
1626 F2
Moderate
Hanoi Flu Pandemic Fairly severe pandemic H5N1
reassortment - originating in Southeast Asia,
Similar to 1918 virus characteristics
Rapid spread R02.75
Moderately high Initial DpC 2.5
Cytokine storm age mortality profile
655 E4
Severe
Niger Virus Very severe pandemic Highly
pathogenic H5N1 reassortment. Unprecedented but
feasible
V. Rapid spread R03.25
Very high Initial DpC 10
Cytokine storm Children infected but not sick
1217 G6
Very Severe
10
Timeline of case loads in different countries
Scenario of Hanoi Flu Event ID 655
Cases per week
15m
SE Asia (excluding China)
10m
USA
Japan
5m
Europe
11
Estimated Impact of National Response Measures
Examples of Efforts to Combat Influenza Virus
with R0 of 2.7
No Intervention
Aggressive Containment
80 CT
90 RP
Tamiflu
80 CT 50 PV
V. Efficient Tamiflu Use
80 CT 70 PV
70 NQ
80 CT 70 NQ
Vaccine
80 CT 50 PV 70 NQ
0
100
of Mortality without Intervention
CT Contact Tracing Targeted Antiviral
Prophylaxis (e.g. Tamiflu) traced RP Ring
Prophylaxis Geographically targeted AP
dosed PV Pre-Vaccination of population with low
efficacy vaccine VE0.5 vaccinated NQ
Neighborhood quarantine confined
12
National Response Measures
Containment and Quarantine
Antiviral Strategy
Primary Healthcare
Tamiflu stockpile, Domestic vaccine production

• United States
• Japan
• Taiwan
• UK
• Thailand
• Malaysia
• China
• South Korea

Deployable mass casualty capability used to
supplement hospitals.
Patient isolation and identification, monitoring,
quarantine of contacts
Target Group Prioritization
Tamiflu stockpile, Domestic vaccine production
Coordination of different levels of medical
supply structure
School closures, work hour reduction, public
transportation closure
Target Group Prioritization
Oseltamivir stockpile
Hospitals activated in accordance with the needs
of flu pandemic.
Isolation, closing schools and restricting public
gatherings
Contribution of 18 m USD for flu vaccine
production
Border control
Tamiflu stockpile, Domestic vaccine production
NHS will establish ways of caring for large
numbers of patients on a scale outside normal
experience,
Public health and/or social distancing measures
to reduce morbidity and/or contain spread
Target Group Prioritization
Increase hospital capacity develop availability
for field hospitals.
Tamiflu stockpile, Domestic vaccine production
Screening of travelers, possible quarantine
measures
Target Group Prioritization
Antiviral stockpile Target Group Prioritization
Quarantine areas within a three km radius of any
suspected case. house to house checks.
Hospital System- Weak System
Antiviral stockpile Domestic Vaccine
Production Target Group Prioritization
All medical resources mobilize and set up
temporary clinics Weak system
Alert System Possible quarantine. Travel
restrictions
Alert System Possible quarantine. Travel
restrictions
Antiviral stockpile Target Group Prioritization
Hospital system
13
Age of insurance portfolios may enhance severe
loss
Age Mortality Profile of Influenza Type
Likelihood of flu type with severity of pandemic
4.0
1
0.9
Normal Flu
0.8
Relative to average in population
3.0
0.7
0.6
0.5
2.0
Cytokine Storm
0.4
0.3
0.2
1.0
Flat
0.1
0
Very Severe
0.0
Slow
lt10
11-17
18-44
45-64
65
Severity of Pandemic
mortality
mortality
mortality
mortality
mortality
Age Distribution of Insured Portfolios

• Cytokine storm kills young adults with strong
  immune response systems
• Cytokine storm is more likely in severe
  pandemics with a pathogenic virus
• Insurance portfolios may have greater
  concentrations of young adults
• More severe pandemics will cause higher losses to
  younger populations

100
65
90
80
45-64
70
18-44
60
11-17
50
0-10
40
30
20
10
0
US Insured
US General Pop
14
Model Predicts Likelihood of Mortality in a
Country
United States Mortality, Year T0 of Pandemic,
F0.03
0.03
0.02
Historical Pandemics, Normalized to modern
population
Prob of Exceedance
RMS Model Event set of 1,890 pandemics
0.01
0.00
0.0
1.0
2.0
3.0
4.0
5.0
Millions
Fatalities in US General Population
15
Key Response Variables Affecting Pandemic Loss

• Vaccine production speed, efficacy and
  manufacturing capacity
• Vaccination implementation
• Behavior of individuals
• Resources and initiative applied by government
  responders
• Time before pandemic arrival
• Moderately sensitive to
• Tamiflu stockpiles
• Primary healthcare quality
• Warning (Disease surveillance capability)
• Less sensitive to
• Location of initial outbreak
• Closure of borders and imposed travel restraints

16
Anti-Viral Stockpiles the next few years
Canada
United States
United Kingdom
Japan
Germany
Canada
Courses as of Total Population
United Kingdom
United States
Japan
Germany
Millions of courses of Tamiflu, as of total
population

• US (300m) UK (60m) Japan (127m)
• 2005 30 10 2.5 4
• 2006 51 17 15 11
• 2007 14.6 24 25 20
• 2008 75 25

17
Understanding the Risk to Your Portfolio
Breakdown of risk by country, cedant, and other
portfolio characteristics
0.03
EP Loss by Country Total 3yrs of Pandemic
United States
Japan
Taiwan
0.02
Hong Kong
Prob of Exceedance
Singapore
United Kingdom
Korea
Thailand
Malaysia
0.01
China
0
Billions
Total Loss to Life Insurance (indivgroupcredit
) over 3 yrs
18
RMS Pandemic Risk Management Products

• Infectious Disease Model
• Desktop software model providing exceedance
  probability output for portfolio-specific data
• Similar structure and analytical framework as
  other RMS catastrophe models
• Uses EDM and RDMs, specifically for life health
  exposures
• Pandemic Alert Service
• Daily monitoring service of clinical data and
  preparedness measures
• Provides early warning of likely pandemic
• Provide estimates of likely losses during a
  pandemic
• Consultancy projects and analytics
• Catastrophe bond modeling
• Development of pandemic investment strategy
• Other potential disease outbreaks
• Currently in development

19
Impact of Pandemic on Non-Life Insurance Lines

• Property claims severity increases
• Building maintenance neglected with absenteeism
• Fire severity increases in unoccupied buildings,
  fire services unmanned
• Increased theft Arson Civil unrest?
• Contingent or Civil Authority Business
  Interruption?
• Liability
• General Negligence of employees by business
  managers public policymakers etc.
• DO Claims from shareholders because of poor
  contingency plans
• Product Claims against prevention equipment
  manufacturers, vaccines
• Cargo
• Goods refused transportation network disrupted
• Travel
• Cancellation Illness suffered abroad
• Auto
• Reduction of accident-related loss from reduced
  traffic
• Mortgage loan default
• Significant increase in mortgage and loan
  defaults,

New Orleans, Hurricane Katrina Unoccupied
premises suffered severe deterioration from
unattended minor water leaks (two weeks in high
humidity)